ADR-010: Deterministic Star-Wars-Style Asset Intake Pipeline

Status: Proposed Date: 2026-03-11 Deciders: Agent Org

Context

The current mod workflow can ingest assets through ad-hoc manual steps, but a repeatable low-poly conversion for a Star Wars conversion requires:

The immediate need is a pre-implementation contract for an assetctl toolchain, not a full production asset conversion subsystem.

Add an explicit agent-facing intake pipeline with machine-readable schemas and policy files.
Add assetctl as a dedicated CLI surface for search, intake, normalize, validate, stylize, register, export-unity.
Use a source-tier model (Sketchfab primary, BlendSwap secondary, ModDB reference, browser fallback last).
Track two independent state dimensions on each asset:
- technical_status (discovered → normalized → validated → ready_for_prototype),
- ip_status (generic_safe vs high_risk_do_not_ship).
Require a manifest conforming to schemas/asset-manifest.schema.json before promotion beyond prototype.

Legal/provenance decisions become auditable and machine-enforced instead of informal.
Asset quality variance is controlled through explicit technical gates and validation reports.
Source/API strategy is configurable without code changes via policy documents.
Pre-implementation artifacts remain non-invasive: no runtime coupling before design freeze.
Future release-safe workflows can be added by changing policy and status transitions, not by rewriting the whole pipeline.

One-off manual asset import:
- rejected due to non-repeatability and weak provenance control.
Full external DAM integration:
- rejected in V1 due to complexity and scope, but preserved as a future extension.
Browser automation as default:
- rejected because it is brittle and harder to audit for policy and retries.